What I uncovered about the breakdown
I wouldn't dismiss this as a one-off glitch. According to user reports, the pattern repeats: after updating, restarting, or importing an agent in Codex Desktop, projects and threads disappear, and then part of the history might suddenly return after reconnecting the folder.
For me, this is an immediate red flag for AI automation within the IDE. If a tool keeps the working context in a local state and then loses its connection to the project, the entire AI development chain turns into a lottery.
The most useful clue from the reports is this: the first thing I would do is re-add the original project folder to Codex and check the local .codex directory in the project or home folder. Several users note that old threads reappeared after doing this.
Another interesting symptom: people don't just lose chats; unread badges reset, and message sorting changes. This looks less like data deletion and more like a flawed resync or reindexing of the local state after an update.
I haven't seen an official public statement about a complete fix yet. There are community discussions and issue reports, but no final answer like «fixed in version X.Y.Z».
If you're experiencing this bug right now, I wouldn't experiment blindly: first, I'd make a backup copy of .codex, then reconnect the project, and only after that consider downgrading to a previous version. Otherwise, you might destroy whatever is still left on the disk.
What this means for business and automation
The problem here isn't just the inconvenience. If you build your AI implementation around an IDE, losing chats means losing decisions, prompts, intermediate checks, and task context.
The only winners right now are teams that don't tie their process to a single local history and instead duplicate crucial steps in git, tickets, or external agent memory. The losers are those who relied on the built-in chat as a dependable work log.
I've long viewed such things as an architectural risk, not just a minor UX bug. That's why at Nahornyi AI Lab, we typically design AI solutions for business so that critical context doesn't live in one fragile desktop client.
If your team has already tied its development to Codex and your work history has started vanishing, let's break down your process layer by layer. At Nahornyi AI Lab, I can help you build an AI integration so that even during IDE failures, you don't lose time, knowledge, and your team's real work.